Skip to content

Latest commit

 

History

History
33 lines (24 loc) · 1.62 KB

train_eval.md

File metadata and controls

33 lines (24 loc) · 1.62 KB
Raw

Prerequisites

Please ensure you have prepared the environment and the nuScenes dataset.

Train and Test

Train VAD with 8 GPUs

cd /path/to/VAD
conda activate vad
python -m torch.distributed.run --nproc_per_node=8 --master_port=2333 tools/train.py projects/configs/VAD/VAD_base.py --launcher pytorch --deterministic --work-dir path/to/save/outputs

NOTE: We release two types of training configs: the end-to-end configs and the two-stage (stage-1: Perception & Prediction; stage-2: Planning) configs. They should produce similar results. The two-stage configs are recommended because you can just train the stage-1 model once and use it as a pre-train model for stage-2.

Eval VAD with 1 GPU

cd /path/to/VAD
conda activate vad
CUDA_VISIBLE_DEVICES=0 python tools/test.py projects/configs/VAD/VAD_base.py /path/to/ckpt.pth --launcher none --eval bbox --tmpdir tmp

NOTE: Using distributed mode (multi GPUs) for evaluation will lead to inaccurate results, so make sure to use non-distributed mode (1 GPU) for evaluation.

Reproduce results with pre-trained weights

If you want to reproduce results with pre-trained weights, please change the img_norm_cfg setting in your config file to following:

img_norm_cfg = dict(
   mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], to_rgb=False)

this is the original setting when we train the model, but we have update it in the recent commit according to this issue's advice. If you use the new img_norm_cfg config, you will get wrong metric results and visualizations.